mylinreg-pkg 1.0.0

A from-scratch ML package for house price prediction

mylinreg-pkg

A from-scratch machine learning package for house price prediction, implementing OLS, Ridge, Lasso, KNN, and Perceptron — built with only NumPy, Pandas, and Matplotlib.

Installation

pip install mylinreg-pkg

Features

Module	Class	Algorithm
linear_model	LinearRegressionOLS	Ordinary Least Squares
ridge	RidgeRegression	Ridge (L2 regularisation)
lasso	LassoRegression	Lasso (L1 coordinate descent)
preprocessing	Preprocessing	Missing values, outliers, standardisation
metrics	Metrics	MAE, MSE, R²
feature_selection	FeatureSelection	Forward selection, backward elimination
diagnostics	Diagnostics	Multicollinearity, residuals
visualization	Visualization	Actual vs predicted, residual plots

Quick Start

import numpy as np
import pandas as pd
from mylinreg_pkg import LinearRegressionOLS, RidgeRegression, LassoRegression
from mylinreg_pkg import Metrics, Preprocessing

# Load and preprocess data
df = pd.read_csv("dataset.csv")
df = Preprocessing.missing_values(df)

# Prepare features and target
X = df[["bhk", "area_sqft", "price_per_sqft"]].to_numpy(dtype=float)
y = df["price_lakhs"].to_numpy(dtype=float)

# Train/test split
split = int(len(X) * 0.8)
X_train, X_test = X[:split], X[split:]
y_train, y_test = y[:split], y[split:]

# Normalise
mean, std = X_train.mean(0), X_train.std(0)
std[std == 0] = 1
X_train = (X_train - mean) / std
X_test  = (X_test  - mean) / std

# OLS
ols = LinearRegressionOLS()
ols.fit(X_train, y_train)
y_pred = ols.predict(X_test)
print("OLS  R²:", round(Metrics.r2(y_test, y_pred), 4))

# Ridge
ridge = RidgeRegression(alpha=1.0)
ridge.fit(X_train, y_train)
y_pred = ridge.predict(X_test)
print("Ridge R²:", round(Metrics.r2(y_test, y_pred), 4))

# Lasso
lasso = LassoRegression(alpha=10.0, iterations=1000)
lasso.fit(X_train, y_train)
y_pred = lasso.predict(X_test)
print("Lasso R²:", round(Metrics.r2(y_test, y_pred), 4))

Algorithms

OLS Regression

Closed-form solution: β = (XᵀX)⁻¹ Xᵀy using Moore-Penrose pseudo-inverse for numerical stability.

Ridge Regression

L2-penalised: β = (XᵀX + αI)⁻¹ Xᵀy — shrinks coefficients to reduce variance.

Lasso Regression

L1-penalised via coordinate descent with soft-thresholding — can zero out coefficients for feature selection.

KNN Classifier (standalone knn.py)

Euclidean distance with min-max normalisation. Use odd k to avoid tie votes.

Perceptron (standalone perceptron.py)

Single-layer binary classifier with Heaviside activation and online weight updates.

Results on Pune Housing Dataset (N=150)

Model	MAE	RMSE	R²
OLS	7,957	11,002	0.6844
Ridge (α=1.0)	7,757	10,678	0.7027
Lasso (α=10.0)	7,953	10,993	0.6849

KNN BHK classification accuracy: 73.3% at k=3.

License

MIT